UBS just mapped out the AI infrastructure build — and the numbers are staggering.
Hyperscaler capex held steady under $160B/year from 2020–2023. Then it broke loose.
UBS projects:
• $283B in 2024
• $492B in 2025
• $1.0T in 2026
• $1.4T in 2027
• $1.6T in 2028
Amazon, Google, Microsoft, and Meta are leading. Oracle, neoclouds, and China Cloud add volume on top.
The three years from 2026–2028 alone: ~$4.1 trillion. That's more than triple the prior six years combined.
Every incremental cloud dollar is now being funneled into AI data centers.
This isn't a typical capex cycle. It's a structural shift in capital intensity — and it's rewriting the playbook for how Big Tech allocates resources.
If you're tracking AI infrastructure plays, supplier leverage, or the next wave of semiconductor demand, this is the baseline assumption you need to work from.
Hyperscaler capex held steady under $160B/year from 2020–2023. Then it broke loose.
UBS projects:
• $283B in 2024
• $492B in 2025
• $1.0T in 2026
• $1.4T in 2027
• $1.6T in 2028
Amazon, Google, Microsoft, and Meta are leading. Oracle, neoclouds, and China Cloud add volume on top.
The three years from 2026–2028 alone: ~$4.1 trillion. That's more than triple the prior six years combined.
Every incremental cloud dollar is now being funneled into AI data centers.
This isn't a typical capex cycle. It's a structural shift in capital intensity — and it's rewriting the playbook for how Big Tech allocates resources.
If you're tracking AI infrastructure plays, supplier leverage, or the next wave of semiconductor demand, this is the baseline assumption you need to work from.